Breast cancer survival predicted by computer model

A new computer model may help researchers better predict how long breast cancer patients will live, a new study suggests.

The model, which was developed in a researchers' contest, uses gene signatures — sets of genes that are all "turned on" at the same time in a patient's cancer — to estimate how long patients will live.

These signatures are actually present in many types of cancer, and the researchers identified them in earlier work. The genes in the signatures play roles in cancer growth, the ability of the cancer to invade tissues and the ability of the immune system to kill cancer cells, said study researcher Dimitris Anastassiou, director of Columbia University's Genomic Information Systems Laboratory.

To test how well the model worked for patients with breast cancer, the researchers entered it in a contest called the Sage Bionetworks/DREAM Breast Cancer Prognosis Challenge. The goal of the contest, which included 350 teams, was to build a model that could accurately predict breast cancer survival. The online contest employed crowdsourcing — teams could borrow ideas from each other to improve their models.

Researchers first developed their model using genetic and clinical information from about 2,000 women with breast cancer. They then tested their model on a new data set of 184 women with breast cancer. The researchers did not know how long these patients had lived.

The model by Anastassiou and colleagues worked the best. Given two patients, their model could predict which one would live longer 76 percent of the time. It also ranked patients based on how long it thought they would survive.

With more refinement, Anastassiou said the model could help researchers decide how to treat individual patients.

Anastassiou said he wants to conduct a study to see if the gene signatures in his model can be combined with, or replace some of the biomarkers researchers use today to better predict survival. Right now, the model is not ready to be used by doctors or patients.

The study is published on April 17 in the journal Science Translational Medicine.

Copyright 2013 MyHealthNewsDaily, a TechMediaNetwork company. All rights reserved. This material may not be published, broadcast, rewritten or redistributed.